QuoteBench: How Matched Scores Can Hide Command-Path Failures

  • 类型:arxiv
  • 标识:2608.13547
  • 链接:https://arxiv.org/abs/2608.13547
  • 主分类:agent
  • 形态:method
  • 被引:0
  • 被引来源:Semantic Scholar + OpenAlex
  • S2被引:0
  • OpenAlex被引:0
  • 影响力被引:0
  • TLDR:Evaluations of command-issuing agents should report the model configuration, generation contract, execution path, operating point, and final-state validator rather than treat a matched score as an intrinsic model property.
  • OpenAlex ID:W7203479664
  • OpenAlex DOI:10.48550/arxiv.2608.13547
  • DOI:10.48550/arxiv.2608.13547
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://arxiv.org/pdf/2608.13547
  • OpenAlex更新:2026-08-31
  • 待LLM分类:否
  • 标题中文:QuoteBench:匹配得分如何掩盖命令路径失败
  • TLDR中文:对指令下发 Agent 的评估应报告模型配置、生成契约、执行路径、工作点及终态校验器,而不应将匹配分数视为模型的内在属性。
  • 来源文件:
  • /inbox/tom/_candidates/2026-08-22-agent-rag-longcontext-candidates.json
  • /inbox/tom/_candidates/2026-08-23-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • [OpenAlex backfill]